00:10Hello and good evening. I'm Melissa Idris. Welcome to Consider This.
00:14This is the show where we want you to consider and then reconsider what you know of the news of
00:19the day.
00:19The artificial intelligence company Anthropic recently revealed how Claude AI was used in a political influence campaign targeting Malaysia.
00:29Now, it involved around 1,000 fake social media accounts, a bogus news site and fabricated intelligence dossiers.
00:38So, just how vulnerable are we as a country to AI-enabled election manipulation?
00:44And what needs to be done to address those vulnerabilities before GE16?
00:49Well, joining me on the show today, I have Munira Mustafa, who is the founder and executive director of Chasa
00:55Group,
00:55which is a research and analytical consulting firm specialising in security challenges.
01:02Munira, thank you so much for joining me on the show today.
01:04So, let's talk about the recent news around Anthropic and the disrupted political influence operation.
01:14So, talk to me from a threat intelligence perspective.
01:19How do you see that? What makes this operation significant?
01:24This is a good example of cognitive warfare because the operation is all about trying to influence and manipulate the
01:31cognitive function psychologically.
01:34It's all about influencing and the intended outcome is to influence behaviour.
01:39And therefore, in that sense, it means devoted choices.
01:43Okay. Well, when you talk about influence behaviour, so Anthropic said it received a request or there was a request
01:50for 1 million artificial views in support of the sitting Prime Minister.
01:54But we don't know whether that means that the views were actually delivered or whether the Prime Minister's office was
01:59involved.
02:00None of that has been confirmed.
02:01But when you talk about influencing behaviour, Munira, how does that work?
02:06How does fake engagement distort perceptions of public opinion or political support?
02:15How does that work and what do you think are the implications of that?
02:18It's about manipulating and manufacturing their reality.
02:22So, if they could create the perception that this candidate is favourable or disfavourable, it would influence the target audience
02:31and it might influence their opinion on the candidate of interest.
02:38And that would have a whole implications on their views and their opinions and that would also influence their ballot
02:48as well.
02:48Okay. So, Anthropic identified some links to an Istanbul company.
02:55It described this company as a paid influence as a service platform.
03:00Is this something that's typical or common that's happening in the industry or happening worldwide?
03:06I had never heard of this before, but from your work, Munira, are you seeing more of this happen?
03:11And are there other tech companies also picking up on this?
03:17Unfortunately, that is what is happening globally.
03:20It's not just Malaysia, it's worldwide.
03:22There are many firms that offer such services.
03:26And it doesn't really reflect their political opinion, but it's more of capitalising on potential opportunities to affect and influence
03:36the political deliberation of the target country.
03:40Because it's very lucrative.
03:41And it also offers possible deniability as well.
03:44So, even if they claim that they're acting in favour of the sitting primary, for example, I mean, that's their
03:53claim.
03:54And right now, you know, like they are also denying their role, their participation in this.
03:59So, that's really, the attribution is always the hardest report, then, you know, there's a high likelihood that that is
04:05what's happening.
04:06But, you know, we have to be cognizant that it's not just this one company.
04:10There are many, many threat actors out there seeing elections.
04:15Because elections are always very, very lucrative for such manipulative operations.
04:21So, we've always had cyber troopers before.
04:25We've always had political influence campaigns and operations before.
04:29But how does AI change the nature, the scale, the economics of influence operations?
04:38Because it expands the operations.
04:41Before this, you have to rely, you have to pay individuals, people, and you can only have about, like, 50,
04:47100 people or so in payroll.
04:48But when you have AI, I mean, the possibilities are endless.
04:51You can create multiples and multiples of accounts and generate many individuals, you know, and cultivate them and craft that
05:00personalities or so-called influencers in today's economy.
05:05And, you know, like, I could literally open any LinkedIn profile and I'm bombarded with so many, very clearly, AI
05:14-generated profiles.
05:15So, you know, it's a minefield out there, basically.
05:20I mean, even the most ordinary things, because they're plentiful out there.
05:28It's not just about politics.
05:30It's also the economic factor as well, because it's a very lucrative industry.
05:34So, it's not, you know, it's not just politics.
05:37They're also literally on YouTube.
05:39They're out there trying to influence your opinion and trying to sell you something or potentially manipulate your opinion so
05:47that the targeted outcome is your response.
05:50your behavior, basically, into compliance.
05:54Okay.
05:55So, a couple of things that you mentioned that I want to delve into.
05:59So, when you say, Anthropic said they've disrupted this operation.
06:03They've removed the clawed account that was involved.
06:05Does that mean that's the end of the threat?
06:09I mean, you're shaking your head, so I'm just curious to know.
06:11How quickly can operations like this come up?
06:15And are we looking at other AI models, other accounts, other platforms that could be used?
06:22I think there's always the kind of, how to say, there's always a gap that these threat actors are looking
06:30to exploit.
06:31So, even if Anthropic already said that they've already removed and expelled the threat actor, they could always come back,
06:38create a different account or under different individual name.
06:41Because that's the thing.
06:42It's easily manufacturable.
06:45And on top of that, saying so, you know, Anthropic can have these guardrails.
06:51But if you've been following the latest news about Anthropic and how it's not been very good at curtailing these
06:57kinds of activities, for example,
07:00I think Tumblr Ridge is one good example of it, if you're aware of the school shooting of Tumblr Ridge
07:08and how the school attacker managed to circumvent all the guardrails that Anthropic has already introduced in the system.
07:17Because, you know, Anthropic doesn't think of the potential loophole that the threat actor could exploit.
07:25Because it's all about trying to work your way around this.
07:28Like, if you post certain questions.
07:30So, for example, Anthropic already set out these parameters.
07:33Like, if a customer asks this question, then the AI should just say, no, we can't do that.
07:41But, again, it's all about manipulation of the words.
07:44Like, trying to play around the words and trying to work around it.
07:47Trying to adapt, basically.
07:49So, you know, you can set out.
07:52It's like a game of chess checkers, for example.
07:55Like, we, as the opposite end, we're thinking of, we're setting out the ground rules like this.
08:02But threat actors don't play by our rules.
08:05They are looking to break our rules.
08:07Right.
08:08Okay.
08:08So, can I circle back to something you said a bit earlier about plausible deniability?
08:16How difficult is it?
08:17So, Anthropic says it hasn't identified the client behind that Malaysia-focused activity behind the Istanbul company.
08:23But how difficult is it, Munira, to identify who is behind these influence operations?
08:30Are you curious to know who might be behind this?
08:35Well, they've already identified the Turkish company behind it, which, obviously, that they've denied.
08:40But they could easily just be the third person or the broker.
08:46Because, you know, there's plenty of players in between.
08:49So, when I say about plausible deniability, it's, I mean, this has always been the puzzle, you know,
08:55that confronts the cybersecurity aspect of things because, you know, trying to attribute who's actually the players,
09:03the people who are actually financing these kind of operations.
09:08You know, you have the financiers, you have the decision makers,
09:11and then you have the actual people in the middle of the field doing all the deployment.
09:18But, you know, one of the ways that you could actually track them down is, obviously, through financial means, for
09:25example.
09:25I always think that trying to track bad actors through financial channels, it's always, you know, it's always a good
09:32idea.
09:32But with that saying, so, even now, it's also still, you know, you're mired with challenges
09:37because they don't pay by the traditional payment system.
09:41They use bitcoins, for example, crypto, and all sorts of various ways of paying all these.
09:49So, you get, trying to, you have to track the activities, and then you have to complement that by identifying
09:54who's the financing,
09:55who's the backers, and everything.
09:57So, that's why the attribution part is always very difficult.
10:00Right.
10:00Okay.
10:00So, what do you, what would you like to see authorities do now?
10:06Where, you know, leading up to the next general elections,
10:09what are you most concerned about, and what would you like to see being addressed as we lead, head to
10:16GE16?
10:17I think people should be able to have all the information available to them.
10:27I mean, I'm, because the trouble is that you have the so-called, you know, you have truth,
10:32and then you have a bunch of manufactured information.
10:35And then, what makes it difficult is that the truth is being baked by all the misinformation altogether.
10:42So, trying to make your, trying to form your opinion is, again, it's a minefield.
10:48Because, you know, there's a lot of, it's peppered with misinformation, and it can be very confusing.
10:54But, I'm still a firm believer that I think that people should have all information available to them,
11:00because they should be able to make the judgment for themselves.
11:03But, you know, in those terms, like, again, this is where I feel that transparency is the key element here.
11:12Because, if you are not going to treat your citizens with that same kind of trust that you expect out
11:19of them.
11:20Munira, thank you so much for joining us on the show.
11:22That was Munira Mustafa from Chasa Group.
11:25And we're going to take a quick break here on Consider This.
11:28We'll be back with more.
11:29Stay tuned.
